Hello Everyone,
Could I query something?
In Cox regression if we are deciding whether or not to stratify, it is recommended in the SPSS manual that a log(-log (Survival)) plot against time is constructed, one plot for each level of the strata.
For example, say that we had variables hair colour (levels: brown and fair) and gender. SPSS constructs the log(-log(Survival)) plot using the baseline survival functions *calculated from the Cox regression* for a stated stratification variable (say gender)
My question is, is this the correct way to assess whether or not we should stratify? Following the text by Marubini and Valsecchi, 1995 (p. 207) shouldnt we instead construct a plot of log(-log(Survival)) against time where the Survival is estimated from the data as 4 *Kaplan Meier Curves* (2 for males (where one curve is for males with brown hair and one curve for males with fair hair) and one for females (one curve for females with brown hair and one curve for females with fair hair)? If so, I guess the correct way to do this is to divide the data into 4 using the following criteria:
Gender Hair group
Male fair 1
Female fair 2
Male brown 3
Female brown 4
And carry out a Cox regression where group is the stratification variable (and there are no covariates)? ....then we can observe the shapes of the resulting curves and their distances apart.
Many thanks for your help in understanding this, Best Wishes, Kim
Send instant messages to your online friends
http://uk.messenger.yahoo.com
Locked
